Dobeu-tech-eco

dt-feature

"Add a feature to the roadmap and update every plan to match: scores clarity, interviews when unclear, has the lead write it into the end state with acceptance criteria, and opens a plan PR. Use for add a feature or put this on the roadmap."

Dobeu-tech-eco 0 Updated 1w ago
GitHub

Install

npx skillscat add dobeu-tech-eco/dts2-marketplace/dt-feature

Install via the SkillsCat registry.

SKILL.md

Cursor procedure

Find the plugin root: the nearest ancestor of this file that contains .cursor-plugin/plugin.json. The examples below say $PLUGIN for that directory.

Run python3 "$PLUGIN/scripts/dt_gate.py" for every gate. Exit 2 from scaffold, preflight, or approve stops the command. Exit 2 from bridge that prints BLOCKED unsupported-runtime-script is a precise blocker for that one step: record it and do not imitate the script. The interview, approve, and scaffold still run.

--yes, --ship, and --scaffold-only are rejected. A JSON key with the same name as a kind is not authorization. approve records the sha256 of the file the user accepted, and later gates require that digest.

Phase files for this command are in $PLUGIN/skills/dt-feature/.

Merges, fleet broadcast, hook install, the headless SDK, and paid provider calls are CURSOR_UNSUPPORTED even after a permission phrase is recorded.

Cursor edition: run scripts/dt_gate.py from this plugin for every gate. Do not read tokens or credential files. Do not merge, broadcast, install hooks, or spend a provider API. Carry out file work with Read, Edit, Grep, Glob, and git.

Feature — from idea to planned sections, every plan in sync

/dt-feature <description> plans a feature; it does not build it. Building is /dt-run section <#> or /dt-autopilot <#…> afterwards.

1. Pre-check and scope

  1. Run the run skill's Pre-check (every run command) exactly as written there. An upgrade it starts stops here; the handoff records Next action: /dt-feature <description>.
  2. python3 "$PLUGIN/scripts/dt_gate.py" bridge scope-check.py . → use the workspace root when it says workspace-member. No .harness/end-state.md → say No end state yet; /dt-setup interviews to one. and stop: a feature needs an approved end state to land in.
  3. Read .harness/end-state.md (Goal, Constraints, Non-goals, Blueprint, Sections, Acceptance criteria, Open questions), .harness/config.yaml paths.docs, .harness/action-items.md, and every plan-like doc the repo keeps (from paths.docs, plus any of PLAN.md, ROADMAP.md, STATUS.md, BRAINSTORM.md, docs/roadmap*, docs/**/plans/* that exist). Record which ones exist; they are the plans to keep in sync.
  4. Conflict check. The feature contradicts a Non-goal or Constraint → quote it and ask once: change the end state (the feature overrides it) or drop the feature. It duplicates an existing component or section → say which and ask: extend that one, or stop.

2. Clarity, then the deep interview only when needed

  1. Rate the feature as one component (or several, when it plainly spans layers) on the setup skill's Round 4 dimensions (goal, constraints, success, and context for a brownfield repo), from the description and the repo only. Do not inflate. Write {"mode":"<greenfield|brownfield>","components":[…]} to a temp file outside the repo and run python3 "$PLUGIN/scripts/dt_gate.py" bridge clarity-score.py <file>.
  2. Threshold: end_state_threshold from .claude/dts2-harness.local.md, default 80. Start the message with Clarity: NN%.
  3. At or above the threshold → go to step 3 (no interview).
  4. Below it → run the deep interview scoped to this feature, using the setup skill's Part 1 rules verbatim, limited to the new component(s):
    • Round 3b. Sections for the feature: each section with files / area, depends on (existing section numbers allowed), complexity, status todo; a section with no files / area or depends on caps clarity exactly as Round 3b says.
    • Round 4 — Question loop, one question per round at weakest.component / weakest.dimension, until the threshold, the literal Continue anyway, or 20 rounds.
    • Round 4b — Skills for the new component only.
    • Round 4c. Grill the draft with /grill-me on the feature's draft (component, Done criteria, Blueprint lines, Sections).
  5. Every answer goes into the record; anything unresolved goes to Open questions with the feature's name.

3. Lead writes the plan change

Dispatch dt:lead (seat: sign-off, model: opus) with the feature record, the current end state, .harness/plan-style.md if present, and the list of plan docs. Tell it to return, as one patch, without touching code:

  • end-state.md: the component under Topology; its - [ ] acceptance criteria under Acceptance criteria; Blueprint rows changed or added with status; new rows in ## Sections numbered after the last existing one, with depends on pointing at real section numbers; Constraints, Non-goals, Assumptions and Open questions as the interview settled them; Skills rows for new skills. Existing sections and approved criteria are never renumbered, rewritten or removed.
  • Each repo plan doc that exists: the feature added in that doc's own format and place (a roadmap entry, a phase task, a status line planned), never a reformat of the rest.
  • action-items.md: a Decisions row for each open question, a Secrets row (name only, via action-items.py <repo> add Secrets NAME) for each new secret the feature needs, a Deploy row for each external setup step.
  • ## Done for the feature: one checkable outcome per line.
    A lead report whose first line is PARTIAL: hit turn cap at <step> is not a patch: do not show it for approval; re-dispatch the lead with a narrower scope or a fresh agent, inside the fix-loop cap.

4. Architecture and agent-friendly passes, then approval

  1. 5b. Architecture pass. Run /improve-codebase-architecture against the repo with the new sections as the planned change; present its findings as amendments to the patch and ask Apply these architecture amendments? (y/n/edit).
  2. Show the patch as a diff summary (end state, each plan doc, action items), the new Sections rows, the Done list and the clarity, and ask once: Add this feature to the plan? Reply "approve", or tell me what to change. On changes, revise with the lead and ask again. Silence means stop.
  3. 5c. Agent-friendly rewrite. After approval, run /writing-for-agents on the new end-state text only; it changes form, never facts; apply after one yes.
  4. In autopilot, steps 2 and 3's questions are answered by the autopilot consensus gate instead of the user (see the autopilot skill).

5. Land it

  1. Branch feature/<slug> from origin/<base> (dev when it exists, else the default branch), write the approved patch, keep status: approved on the end state with a - <date> feature <slug> added: sections <#…> line in .harness/state.md → ## Decisions.
  2. Ship it: run /dt-ship (its pre-check, checks, receipt and PR); the PR's Linked work names the feature slug and the new section numbers.
  3. Refresh the handoff: python3 "$PLUGIN/scripts/dt_gate.py" bridge handoff-write.py <repo>.
  4. End with one line naming what to run next: /dt-run section <first new #> or /dt-autopilot <#,#,…>.

Guardrails

  • Planning only: no code, no migrations, no installs beyond the companion skills.
  • Never renumber, rewrite or delete approved sections or criteria; only add, or change what the user explicitly approved.
  • Questions one at a time, never more than one per message.
  • Never print a secret's value; secrets are names only.